Define Fibonacci Arcs before you test them
Fibonacci Arcs are a drawing method that uses Fibonacci-derived angles or distance relationships to project curved levels from a chosen reference point to later points on a price chart. A responsible backtest starts by stating the exact mechanics you will evaluate, because different interpretations of arcs can produce different levels.
Write down, in plain terms, what you are computing. For example:
- What are the reference points (swing high/low, start/end bar)?
- How are Fibonacci levels converted into arcs (which ratio set, which transformation)?
- Which timeframe and price type are used (close, high/low, or mid)?
- How do you handle updates (re-draw on every new bar, or only when a new pivot forms)?
- How do you map arc geometry to measurable outcomes (distance-to-arc, touch/cross conditions, or rule-based triggers)?
If you cannot precisely define these items, you cannot compare backtest runs or explain results to others.
Separate stable indicator mechanics from variable conditions
Backtesting combines two different things: the stable “how the arc is drawn” logic, and variable market/provider conditions that affect realized outcomes.
Keep these layers separate in your workflow:
- Stable mechanics (fixed): the arc construction rule, the measurement of arc interaction, and the decision logic you test.
- Variable conditions (scenario assumptions): the dataset, data granularity, execution model, and transaction costs.
This separation matters because a good-looking result can come from favorable assumptions rather than the arc logic. If two researchers test the same arc mechanics but use different costs, execution timing, or pivot-detection rules, their results may not be comparable.
Set assumptions for data, costs, and execution
A responsible backtest does not only ask “what happened historically?” It also asks “what would have been possible given realistic constraints?”
Define the following as explicit assumptions:
- Data: what instrument(s), what time range, and what bar size. Also define whether you use adjusted data (for corporate actions) if relevant.
- Sampling and alignment: confirm that the arc is computed using only information available at the decision time. A common failure mode is accidentally using future information when determining pivots.
- Costs (cost types): model at least one cost component, such as spread/slippage or a generic per-trade cost. “No costs” can overstate performance.
- Execution timing: when does a position enter and exit relative to the bar that triggered the event? For example, is it at the bar open of the next bar, or at the same bar close?
These assumptions are the “kostensoorten” of your test: they describe which frictions you included and which you intentionally left out.
Control bias with parameter locking and multiple datasets
Backtests are prone to bias. You can reduce it by treating the process like a measurement, not a tuning exercise.
Practical bias controls include:
- Parameter locking: decide on arc-related parameters (pivot rule details, thresholds, interaction definitions) before you evaluate performance. Only then run the backtest.
- Walk-forward testing: if you must re-fit parameters, do it sequentially over time, using earlier periods to set parameters and later periods to evaluate.
- Out-of-sample checks: keep at least one portion of the dataset untouched until the final evaluation.
Use “aannames” (assumptions) consistently across all runs. If changing a detail materially improves results, that detail is a candidate for overfitting—especially if it has been tuned to a single market regime.
Use an evidence test that measures more than one metric
Choose outcome criteria that match the question. For arc research, you might measure:
- How often the arc interaction rule is triggered (frequency)
- Whether subsequent price movement (defined window) behaves differently versus a baseline
- Distributional effects (not only average returns)
Even without recommending trades, you can still define an evidence standard such as: “the arc interaction rule produces a statistically distinguishable effect from a naive baseline across multiple periods.”
Because outcomes vary with “variabele factoren” (market regime, volatility, liquidity, and how pivots form), report where the effect is stable and where it fails.
Limitations and failure modes to expect
A responsible backtest must include material limitations. At least these are common failure modes for chart-pattern and geometric-indicator testing:
- Overfitting: tuning arc parameters to historical quirks that do not repeat. - Data leakage: pivot detection or indicator updates inadvertently use future information.